PARALLEL COMPUTING SCHEME FOR THE ENCRYPTION PROCESS OF DNSCRYPT PROTOCOL USING CUDA

• Main Author: ASTRA PRATAMA (NIM : 13514104), FAIRUZ
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/27087
• Institution: Institut Teknologi Bandung

<p align="justify"> DNS, which provides the internet naming services, has a privacy problem because all communication between client and resolver can be read by a third party. DNSCrypt can be used to solve this problem, by encrypting the communication channel used between the client and the resolver. However, the use of DNSCrypt will add a burden of encryption and decryption of messages sent between clients and resolvers that will negatively impact the system performance. One solution that can be used is to utilize CUDA to perform encryption processing in parallel using the GPU. By processing encryption and decryption in the GPU, processing time of a query can be reduced, improving performance. Implementation of this system can be done by changing the implementation of the DNSCrypt proxy application and adding CUDA usage selection feature to encrypt / decrypt the message. After that, performance measurement and comparison will be performed by measuring the average processing time of a DNS query from the various systems created. <br /> <br /> From the test results, it can be concluded that CUDA can be used in the DNSCrypt protocol to process message encryption in parallel by running multiple CUDA threads in parallel, where each thread will process a single message block / section that can be processed independently. In addition, from the test results, it is found that although a DNS server will have the same performance regardless of the average response size of the DNS query served, the performance of the DNSCrypt system will get worse as the average response size increases. The use of CUDA in the encryption and decryption response process will reduce this effect, but will add an overhead to the processing of each query. In addition, for small data, processing uses only a small portion of the GPU processing cores. Both of these factors make the speed-up algorithm very small for relatively small data. Therefore, CUDA is only suitable for processing the response of queries in the DNSCrypt protocol if the average size of the system response is large enough. <p align="justify">